Gasgoo Munich-ACE ROBOTICS has teamed up with The Chinese University of Hong Kong's Multimedia Laboratory (CUHK MMLab) to release ACE-Ego, a new "one-brain, multi-type" embodied operation VLA model, and open-source it to the industry, Gasgoo learned on June 17.
Image Source: ACE ROBOTICS
On the internationally recognized RoboCasa GR1 TableTop benchmark for humanoid robots, ACE-Ego set a new record with a 72.8% average success rate—claiming the top spot and significantly outperforming mainstream models like Nvidia's GR00T, PI π₀.₅, and JD's JoyAI-RA. Meanwhile, in the rigorous domain randomization tests of the challenging dual-arm manipulation benchmark RoboTwin 2.0, ACE-Ego achieved a 90.62% success rate, demonstrating environmental robustness far above the industry average.
Back in December 2025, ACE ROBOTICS introduced a "Human-centric" ACE embodied research paradigm. By treating the laws of human-physical world interaction as the core starting point, the company built a comprehensive technical framework spanning "environmental data collection — Kaiwu World Model 3.0 — embodied interaction."
ACE-Ego can now reliably handle complex, long-cycle retail tasks involving strong physical contact—such as packing plastic bags and boxing shoes. Covering key offline retail stages like product sorting and order fulfillment, the model breaks through the limitations of earlier systems that were limited to simple tabletop grasping. This offers a cost-effective technical pathway for embodied intelligence to achieve large-scale industrial deployment.
